The rehabilitation robot is a strongly coupled system.Compliant and safe interaction training environment is of great significance to improve the rehabilitation effect on patients.The existing interaction control methods mainly involve movement intention recognition of the human body and interactive control strategies.Movement intention recognition of the human body is generally based on bioelectrical signals or interactive forces/moments, while interactive control strategies include virtual tunnels, impedance control, and functional electrical stimulation.Stable and safe interaction environment is essential to implement rehabilitation training smoothly and avoid secondary injuries to patients.This study fully reviews the above issues and analyzes the existing problems in depth.

In this paper, a robot, namely iLeg, is designed for the purpose of rehabilitation of patients with hemiplegia or paraplegia. The iLeg is composed of one reclining seat and two leg orthoses, and each leg orthosis has three degrees of freedom, which correspond to the hip, knee, and ankle. Based on this robotic system, two controllers, i.e., passive training controller and active training controller, are proposed. The former takes advantage of the proportional-integral control method to solve the trajectory tracking problem, and the latter employs the surface electromyography signals to achieve active training. Two simplified impedance controllers, i.e., damping-type velocity controller and spring-type position controller, are designed for active training. A perceptron neural network detects movement intentions. The performance of the controllers was investigated with one able-bodied male. The results showed that the leg orthosis tracked the predefined trajectory based on the passive training controller, with the error rates of 0.45%, 0.44%, and 0.27%, respectively, for the hip, knee, and ankle. The active training controller whose loop rate is 6.67 Hz can move the leg orthosis smoothly, and the average recognition error of the perceptron neural network is less than 5%.

Many rehabilitation robots have been designed to alleviate the conflict between increasing number of post-stroke patients and shortage of therapists. Active training is the main feature of advanced rehabilitation robots, which has been proved to be more effective than simple passive movement training. This paper presents the implementation of active training on a 2-DOF upper-limb rehabilitation robot, which can assist the shoulder and elbow joint rehabilitation training of post-stroke patients. The controller is built based on impedance control, which can provide a compliant human-robot reaction. The implementation of active training is combined with a virtual reality game, and the average error between the actual and target reaction force is 1.41 ± 0.79 N in the X axis, and 1.22 ± 0.91 N in the Y axis.

sEMG (surface electromyography) signals have been used as human-machine interface to control robots or prostheses in recent years. sEMG-based torque estimation is a widely research methodology to obtain human motion intention. Most researches focus on improving the accuracy of sEMG-torque models, which often makes them complicated and confined in the laboratory research. However, an accurate estimation of muscle torque could be unnecessary to perform the robot-assisted rehabilitation training. This paper proposes a practical method to estimate the net muscle torques of lower limbs using sEMG, which can be used to implement a real-time coordinated active training with iLeg-a horizontal exoskeleton for lower limb rehabilitation developed at our laboratory. Two three-layer back propagation (BP) neural networks are built to estimate the net muscle torques at hip and knee joints respectively. Experimental results show that the well-trained neural networks estimate the user's motion intention in real-time, and can assist the user to perform an active training with iLeg.

In this paper, an sEMG-driven musculoskeletal model of human shoulder and elbow joints is built based on time delay neural network (TDNN). Six principal muscles of the upper arm and forearm are included, and the experiment was conducted under isometric contractions with the aid of a planar haptic interface. Both force amplitude and direction were regulated continuously, and the experiment results proved the effectiveness and performance of this modeling method. The model was proved to have less overfitting risk than the most-used basic multilayer forward networks, and the isometric model was proved to be still effective in estimation of slow movement cases.

This paper is a continuation of previous published work by the same authors on Personalized Modelling and Evolving Spiking Neural Network Reservoir architecture (PMeSNNr). The focus is on improvement of predictive modeling methods for the stroke occurrences case study utilizing an enhanced NeuCube architecture. The adaptability of the new architecture leads towards understanding feature correlations that affect the outcome of the study and extracts new knowledge from hidden patterns that reside within the associations. Through this new method, estimation of the earliest time point for stroke prediction is possible. This study also highlighted the improvement from designing a new experimental dataset compared to previous experiments. Comparative experiments were also carried out using conventional machine learning algorithms such as kNN, wkNN, SVM and MLP to prove that our approach can result in much better accuracy level.

To design a control strategy for iLeg, an exoskeleton robot developed for lower limb rehabilitation aiming at investigating the feasibility of integrating functional electrical stimulation (FES) with robot-based rehabilitation training, an FES-assisted training strategy combined with impedance control, has been proposed in this paper. Through impedance control, an active compliance of the robot is established, and the patient’s voluntary effort to accomplish the training task is inspired. During the training process, the patient’s related muscles are applied with FES which provides an extra assistance to the patient. The intensity of the FES is properly chosen in order to induce a desired active torque which is proportional to the voluntary effort extracted from the electromyography signals of the related muscles using back propagation neural networks. This kind of enhancement serves as a positive feedback which reminds the patient of the correct attempt to fulfill the desired motion. FES control is conducted by a combination of neural network-based feedforward controller and a PD feedback controller. Simulation conducted using Matlab and the experiment with a spinal cord injury subject and a healthy subject have shown satisfactory results which verify the feasibility of this control strategy.

The number of paralytic sufferers is currently growing huge and the rehabilitation for them is usually a long-time process. Compared to the traditional physiotherapy, rehabilitation with the assistance of robots can reduce the cost and time, and less labor intensity is required. Moreover, various training strategies are provided by robots, so that rehabilitation effect can be improved. Lower limb rehabilitation robots are categorized into horizontal exercisers, vertical locomotors, sit-to-stand aids and multi-orientation hybrids, according to the posture of patient during therapy. Horizontal exercisers are subcategorized into end effectors and exoskeletons, and vertical locomotors are further grouped as suspending body weight support (sBWS) based gait trainers and stand-alone wearables. Interactive control between mechanism and patient is required to create a secure, comfortable and natural training environment for paralytic patients. According to the signals employed to deduce the movement intention of patients, interactive control methods are classified into force-based control and biomedical-signal-based control. Two approaches that are in particular worth mentioning for force-based interactive control are hybrid force-position control and impedance control. Surface electromyogram (sEMG) and electroencephalogram (EEG) are two mostly used signals for biomedical-signal-based control.

In this paper, surface electromyography (sEMG) from muscles of the lower limb is acquired and processed to estimate the single-joint voluntary motion intention, based on which, two single-joint active training strategies are proposed with iLeg, a horizontal exoskeleton for lower limb rehabilitation newly developed at our laboratory. In damping active training, the joint angular velocity is proportionally controlled by the voluntary effort derived from sEMG, performing as an ideal damper, while spring active training aims to create a spring-like environment where the joint angular displacement from the constant reference is proportionally controlled by the voluntary effort. Experiments are conducted with iLeg and one healthy male subject to validate the feasibility of the two single-joint active training strategies.

The paper is a feasibility analysis of using the recently introduced by one of the authors spiking neural networks architecture NeuCube for modelling and recognition of complex EEG spatio-temporal data related to both physical and intentional (imagined) movements. The preliminary experiments reported in the paper suggest that NeuCube is much more efficient for the task than standard machine learning techniques, resulting in high recognition accuracy, a better adaptability to new data, a better interpretation of the models, leading to a better understanding of the brain data and the processes that generated it.

Repetitive activities of daily living (ADL) and robotic active training are commonly practised in the rehabilitation of paralyzed patients, both of which have been proven rather effective to recover the locomotor function of impaired limbs. ADL classification based on electroencephalogram (EEG) is of great significance to perform active robotic rehabilitation for patients with complete spinal cord injury (SCI) who lose locomotion of affected limbs absolutely, where surface electromyography (sEMG) or active force signal can hardly be detected. It is a challenge to achieve a satisfying result in neuro-rehabilitation robotics using EEG signals due to the high randomness of the EEG data. A classification method is proposed based on spiking neural networks (SNN) to identify the upper-limb ADL of three classes with 14-channel EEG data. The continuous real-number signals are firstly encoded into spike trains through Ben's Spike Algorithm (BSA). The generated spikes are then submitted into a 3-D brain-mapped SNN reservoir called NeuCube trained by Spike Timing Dependant Plasticity (STDP). Spike trains from all neurons of the trained reservoir are finally classified using one version of dynamic evolving spiking neuron networks (deSNN) - deSNNs. Classifications are presented with and without NeuCube respectively on the same EEG data set. Results indicate that using the reservoir improves identification accuracy which turns out pretty promising despite that EEG data is highly noisy, low frequently sampled, and only from 14 channels. The classification technique reveals a great potential for the further implementation of active robotic rehabilitation to the sufferers of complete SCI.

A crus adjustment system for a rehabilitation robot includes a crus extension/contraction unit and a knee joint rotation unit. A motor (15) drives a screw rod (2) to rotate, and accordingly drives a screw rod nut and a connecting frame (4) to cause a dovetail groove guide rail (1) fixed to the connecting frame (4) to move, thus enabling extension/contraction of the crus during rehabilitation training. A pulley (10) drives a torque sensor (25) fixed on a torque sensor seat (24), and accordingly drives the whole knee joint, thus enabling the knee joint to rotate. The system has two degrees of freedom, and has the advantages of high positional precision, good intuitive sense for movement, high overall rigidity, simple structure and easy manufacture.

One of the most important issues among active rehabilitation technique is how to extract the voluntary intention of patient through bio-signals, especially EEG signal. This pilot study investigates the feasibility of utilizing a 3D spiking neural networks-based architecture named NeuCube for EEG data classification in the rehabilitation practice. In this paper, the architecture of the NeuCube is designed and a Functional Electrical Stimulation (FES) rehabilitation scenario is introduced which requires accurate classification of EEG signal to achieve active FES control. Three classes of EEG signals corresponding to three imaginary wrist motions are collected and classified. The NeuCube architecture provides promising classification results, which demonstrates our proposed method is capable of extracting the voluntary intention in the rehabilitation practice.

In order to investigate the feasibility of integrating functional electrical stimulation (FES) with robot-based rehabilitation training, this paper proposes an FES-assisted training strategy combined with impedance control for our self-made exoskeleton lower limb rehabilitation robot. This control strategy is carried out in a leg press task. Through impedance control, an active compliance of the robot is established, and the patient's voluntary effort to accomplish the task is inspired. During the training process, the patient's related muscles are applied with FES which provides an extra assistance to the patient. The intensity of the FES is properly chosen aiming to induce a desired active torque which is proportional to the voluntary effort of the patient. This kind of enhancement serves as a positive feedback which reminds the patient of the correct attempt to fulfill the desired motion. FES control is conducted by a combination of neural network-based feedforward controller and a PD feedback controller. The feasibility of this control strategy has been verified in Matlab.

In this paper, an artificial neural network is proposed to estimate knee joint angle in hybrid activation of knee extension motion, including voluntary muscle contraction and functional electrical stimulation (FES) induced contraction. Voluntary electromyography (EMG) signals of three muscles responsible for knee extension and FES parameter which describe the FES intensity are used as input vector of the neural network, while the estimated knee angle is the output. During the experiment, FES with different combinations of parameters (pulse amplitude and pulse width) was delivered to the rectus femoris muscle of a healthy male subject when the knee was in a periodic extension motion by voluntary muscle contraction. Raw EMG signals of three muscles, parameters of FES as well as the actual knee angle were recorded. Totally, there were 52,233 and 17,420 sampling points corresponding to 261 and 87 seconds used to train and validate the neural network. The result shows the trained network has a satisfactory performance on knee joint angle estimation whose output well follows the curve of actual knee angle. Root mean square error between estimated angle and actual angle is employed to represent the estimation accuracy which is 5.07 degree according to the experimental data.

A vast amount of complex spatio-temporal brain data, such as EEG-, have been accumulated. Technological advances in many disciplines rely on the proper analysis, understanding and utilisation of these data. In order to address this great challenge, the paper utilizes the recently introduced by one of the authors 3D spiking neural network environment called NeuCube for spatio-temporal EEG data classification. A methodology is proposed and illustrated on two small-scale examples: classifying EEG data for music- versus noise perception, and person identification based on music perception. Future development and usage of the NeuCube environment can be expected to significantly further the creation of novel brain-computer interfaces, cognitive robotics and medical engineering devices.

A horizontal exoskeleton for lower limb rehabilitation called iLeg has been developed by our laboratory which consists of two 3-DOF (degrees of freedom) robotic leg orthoses. This paper proposes a position-based impedance control with the compensation of BP NNs (back propagation neural networks) for the exoskeleton. Based on the control scheme, the task-oriented active training is investigated where impedance parameters are self-adjusted to movement deviation and patient activities with fuzzy logic. An adaptive haptic interface of active compliance is ensured to provide positive feedback to patients when their effort is desired or negative otherwise, which encourages patients to practice the desired movement following the predefined directed path. Besides, the timing freedom is separated from spatial trajectory and determined by patients. Voluntary effort hence becomes a requirement during the exercises, no effort no movement, so that active contribution of patients is highly motivated. Simulation results have verified the feasibility of the control scheme and the training strategy. An active compliant environment is created with adaptive haptic interface for task-oriented patient-driven training of multi-joint coordination.

This paper proposes a functional electrical stimulation (FES)-involved control strategy for self-made exoskeleton lower limb rehabilitation robot for the training purpose of paraplegic patients caused by spinal cord injury (SCI) or stroke. Two muscles (Vastus Medialis and Riceps Femoris) are stimulated to produce active torque around knee joint which can be considered as a redundant actuator besides electrical motor. During the predefined trajectory tracking task, electrical motors compensate for the gravitational torque of the entire human-robot system, while the muscles provide torque calculated by a PD position/velocity controller based on the tracking error. The FES-induced torque control is accomplished with combination of feedforward and feedback controller, former of which is obtained by applying off-line trained neural networks to map the relationship between desired active torque and FES parameters. Simulation results obtained by using Simulink toolboxes in Matlab verify the feasibility of this control strategy.